Projects

Selected work across AI, cybersecurity, product strategy, and threat intelligence.

Command

The problem

A security analyst investigating a breach might touch six tools in twenty minutes: a SIEM, a threat intel feed, an endpoint console, a ticketing system, a browser, and a chat window to ask a colleague what happened last time. The problem is not any single tool. The problem is that the work lives in the gaps between them.

What it is

Command is a conversational AI layer that sits across Axur's product ecosystem (Axur has been part of Infoblox since 2026). Instead of moving between dashboards, an analyst describes what they need in natural language: correlate these indicators, pull the context from last week's incident, draft the escalation. The system retrieves, synthesizes, and acts.

My role

I helped shape what this product should be, not just technically but conceptually: what it means for AI to move from isolated question-answering toward coordinated operational judgment. Product vision, experience structure, and the longer-term direction toward agentic orchestration.

Why it matters

The difference between automation and useful AI is the difference between scaling rules and scaling judgment. Most security automation runs playbooks. Command is aimed at the harder layer: the one where context matters and the right next step depends on what just happened.

Axur's CTI Platform

The problem

Raw threat data is abundant. What security teams lack is not information but contextualized intelligence: which signals matter, how threats connect to each other, and what deserves attention before everything else. Without that structure, more data often means more noise.

What it is

A platform that centralizes and contextualizes cyber threat intelligence from hundreds of sources, so analysts can work with structured insight instead of fragmented alerts. Earlier in its internal history, this initiative was known as Polaris.

My role

I conceived the product and helped build it from its earliest direction: defining what it should do, collaborating with engineering and data science on how, and shaping the way its value should be understood by the market.

Why it matters

Threat intelligence becomes strategically useful when it improves judgment, not volume. The platform was built on that premise: transform fragmented signals into something an analyst can actually reason with. It now continues inside Infoblox, which acquired Axur in 2026.

Agentic OS

The problem

Most organizations bolt AI assistants onto workflows designed for a world without them: the org chart, the meeting, the handoff all assume human-only knowledge work. The question is not how to use agents inside that structure, but what structure knowledge work should have once agents are part of the team.

What it is

A working model, and a public manifesto, for restructuring knowledge work around human-agent teams, written with Fábio Ramos out of seven weeks of actual practice rather than speculation. The manifesto, a field report, and a starter kit are public.

My role

Co-author. The essays on this site about agentic work and organizational form are the philosophical side of the same investigation; Agentic OS is where those ideas had to survive contact with a real organization.

Why it matters

Everyone has an opinion about agents and the future of work. Documented practice is rarer: what actually changes when a team runs this way for weeks, what breaks, and which parts of the old structure turn out to have been load-bearing.

Read the manifesto ↗Field report and starter kit (GitHub) ↗